Morgan Stanley sharply raised its capital expenditure forecasts for the world's five largest cloud hyperscalers—Meta, Amazon, Microsoft, Google, and SpaceX—projecting that total capex will approach **$1.4 trillion** by **2028**, driven by soaring hardware costs and data center construction delays. A report led by analyst Brian Nowak raised the **2027** and **2028** capex forecasts by **9%** and **10%** , respectively, reaching approximately **$1.2 trillion** and **$1.4 trillion**. The immediate trigger is a broad increase in hardware costs, with overall GPU-related costs rising by about **20%** . Memory's share of total cost for Nvidia's Blackwell generation racks has climbed from the low single digits to the high single digits, and for next-generation Vera Rubin racks, it is projected to jump to about **25%** . Outside-the-rack costs for GB200, TPU, and Trainium are roughly **$11 million to $13 million per megawatt**, while next-generation GB300, Vera Rubin, and Rubin Ultra will see costs rise to about **$16 million to $19 million per megawatt**. Construction cost per gigawatt ranges from approximately **$20 billion** for Amazon Trainium3 to approximately **$49 billion** for Vera Rubin. The report's core thesis shifted from debating whether capex will increase to exploring how much effective computing power this spending can generate and whether future revenues can